Scholarlink

<p>The second most common cause of mortality for women is breast cancer (BC) following lung cancer. Although manual identification can be laborious and subject to error, early discovery is essential for lowering death rates. This study introduces a Path Enhanced Holographic Convolutional Neural Network with Nadam Optimizer (PEHCNN-NO) for automated BC classification. Mammograms from the MIAS as well as CBIS-DDSM datasets were used for evaluation. In pre-processing, a Trainable Self-Guided Filter (TSGF) was applied to reduce noise and enhance breast region visibility. Segmentation was performed using Automatically Weighted Binary Multi-View Clustering (AW-BMVC), which maps multi-view features into a binary space to highlight regions such as masses and calcifications. For classification, PEHCNN integrates holographic computing and transformer-based attention mechanisms, while the Nadam Optimizer (NO) adjusts parameters to improve convergence. Results from experiments indicate that PEHCNN-NO achieves accuracy of 99.92% (MIAS) and 99.96% (CBIS-DDSM), with corresponding precision of 99.67% and 99.74%, and F1-scores of 99.79% and 99.80%. These findings indicate that PEHCNN-NO provides reliable performance for early BC detection in mammography datasets.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Accelerated Breast Cancer Classification from Mammograms Using Path Enhanced Holographic Convolutional Neural Network with Nadam Optimizer

  • 期刊论文

  • 发表日期：

    2025年12月18日

查看全文
  • Harish Kumar,
  • Sheng-Lung Peng,
  • Rupali Mahajan,
  • Anuradha Taluja

摘要

The second most common cause of mortality for women is breast cancer (BC) following lung cancer. Although manual identification can be laborious and subject to error, early discovery is essential for lowering death rates. This study introduces a Path Enhanced Holographic Convolutional Neural Network with Nadam Optimizer (PEHCNN-NO) for automated BC classification. Mammograms from the MIAS as well as CBIS-DDSM datasets were used for evaluation. In pre-processing, a Trainable Self-Guided Filter (TSGF) was applied to reduce noise and enhance breast region visibility. Segmentation was performed using Automatically Weighted Binary Multi-View Clustering (AW-BMVC), which maps multi-view features into a binary space to highlight regions such as masses and calcifications. For classification, PEHCNN integrates holographic computing and transformer-based attention mechanisms, while the Nadam Optimizer (NO) adjusts parameters to improve convergence. Results from experiments indicate that PEHCNN-NO achieves accuracy of 99.92% (MIAS) and 99.96% (CBIS-DDSM), with corresponding precision of 99.67% and 99.74%, and F1-scores of 99.79% and 99.80%. These findings indicate that PEHCNN-NO provides reliable performance for early BC detection in mammography datasets.